diff --git a/ext/encoders_test.go b/ext/encoders_test.go
index 1d526a15d..22e02c1bf 100644
--- a/ext/encoders_test.go
+++ b/ext/encoders_test.go
@@ -249,8 +249,10 @@ func TestEncodersCosts(t *testing.T) {
 				"x": 100,
 			},
 			estimatedCost: checker.CostEstimate{Min: 2, Max: math.MaxUint64},
-			actualCost:    1,
-			version:       1,
+			// json.encode reports an unbounded actual cost, which saturates the
+			// running total rather than overflowing it back to a small value.
+			actualCost: math.MaxUint64,
+			version:    1,
 		},
 	}
 	for _, tc := range tests {
@@ -337,3 +339,55 @@ func TestJSONEncodeCostUnbounded(t *testing.T) {
 	}
 }
 
+// TestJSONEncodeCostUnboundedWithAccruedCost checks that the unbounded cost of json.encode
+// survives being combined with cost accrued elsewhere in the expression, and that it still
+// trips a cost limit. A total which overflowed back to a small value would not.
+func TestJSONEncodeCostUnboundedWithAccruedCost(t *testing.T) {
+	env, err := cel.NewEnv(Encoders(EncodersVersion(1)), cel.Variable("v", cel.StringType))
+	if err != nil {
+		t.Fatalf("cel.NewEnv() failed: %v", err)
+	}
+	exprs := []string{
+		"json.encode(v)",
+		"json.encode(v) == json.encode(v)",
+		"size(v) > 0 && json.encode(v) != ''",
+	}
+	for _, expr := range exprs {
+		t.Run(expr, func(t *testing.T) {
+			ast, iss := env.Compile(expr)
+			if iss.Err() != nil {
+				t.Fatalf("env.Compile(%q) failed: %v", expr, iss.Err())
+			}
+			est, err := env.EstimateCost(ast, testCostHintEstimator{})
+			if err != nil {
+				t.Fatalf("env.EstimateCost() failed: %v", err)
+			}
+			if est.Max != math.MaxUint64 {
+				t.Errorf("env.EstimateCost() got max %d, wanted %d", est.Max, uint64(math.MaxUint64))
+			}
+
+			prg, err := env.Program(ast, cel.CostTracking(nil))
+			if err != nil {
+				t.Fatalf("env.Program() failed: %v", err)
+			}
+			_, det, err := prg.Eval(map[string]any{"v": "hello"})
+			if err != nil {
+				t.Fatalf("prg.Eval() failed: %v", err)
+			}
+			if det.ActualCost() == nil {
+				t.Fatal("det.ActualCost() got nil, wanted a value")
+			}
+			if *det.ActualCost() != math.MaxUint64 {
+				t.Errorf("det.ActualCost() got %d, wanted %d", *det.ActualCost(), uint64(math.MaxUint64))
+			}
+
+			limited, err := env.Program(ast, cel.CostLimit(1000))
+			if err != nil {
+				t.Fatalf("env.Program() failed: %v", err)
+			}
+			if _, _, err := limited.Eval(map[string]any{"v": "hello"}); err == nil {
+				t.Error("prg.Eval() got nil error, wanted cost limit exceeded")
+			}
+		})
+	}
+}
diff --git a/interpreter/runtimecost.go b/interpreter/runtimecost.go
index 5558cf6fd..960f52a00 100644
--- a/interpreter/runtimecost.go
+++ b/interpreter/runtimecost.go
@@ -108,7 +108,7 @@ func (ct *costTrackerFactory) Observe(vars Activation, id int64, programStep any
 	case ConstantQualifier:
 		// TODO: Push identifiers on to the stack before observing constant qualifiers that apply to them
 		// and enable the below pop. Once enabled this can case can be collapsed into the Qualifier case.
-		tracker.cost++
+		tracker.cost = cost.SafeAdd(tracker.cost, 1)
 	case InterpretableConst:
 		// zero cost
 	case InterpretableAttribute:
@@ -118,7 +118,7 @@ func (ct *costTrackerFactory) Observe(vars Activation, id int64, programStep any
 			tracker.stack.drop(a.falsy.ID(), a.truthy.ID(), a.expr.ID())
 		default:
 			tracker.stack.drop(t.Attr().ID())
-			tracker.cost += common.SelectAndIdentCost
+			tracker.cost = cost.SafeAdd(tracker.cost, common.SelectAndIdentCost)
 		}
 		if !tracker.presenceTestHasCost {
 			if _, isTestOnly := programStep.(*evalTestOnly); isTestOnly {
@@ -150,20 +150,20 @@ func (ct *costTrackerFactory) Observe(vars Activation, id int64, programStep any
 	case *evalFold:
 		tracker.stack.drop(t.iterRange.ID())
 	case Qualifier:
-		tracker.cost++
+		tracker.cost = cost.SafeAdd(tracker.cost, 1)
 	case InterpretableCall:
 		if argVals, ok := tracker.stack.dropArgs(t.Args()); ok {
-			tracker.cost += tracker.costCall(t, argVals, val)
+			tracker.cost = cost.SafeAdd(tracker.cost, tracker.costCall(t, argVals, val))
 		}
 	case InterpretableConstructor:
 		tracker.stack.dropArgs(t.InitVals())
 		switch t.Type() {
 		case types.ListType:
-			tracker.cost += common.ListCreateBaseCost
+			tracker.cost = cost.SafeAdd(tracker.cost, common.ListCreateBaseCost)
 		case types.MapType:
-			tracker.cost += common.MapCreateBaseCost
+			tracker.cost = cost.SafeAdd(tracker.cost, common.MapCreateBaseCost)
 		default:
-			tracker.cost += common.StructCreateBaseCost
+			tracker.cost = cost.SafeAdd(tracker.cost, common.StructCreateBaseCost)
 		}
 	}
 	tracker.stack.push(val, id)
diff --git a/interpreter/runtimecost_test.go b/interpreter/runtimecost_test.go
index b160318b7..e9a4c11e7 100644
--- a/interpreter/runtimecost_test.go
+++ b/interpreter/runtimecost_test.go
@@ -904,3 +904,52 @@ func TestRuntimeCost(t *testing.T) {
 		})
 	}
 }
+
+// TestRuntimeCostUnboundedOverloadSaturates verifies that an overload which reports an unbounded
+// actual cost saturates the running total rather than overflowing it back to a small value. A
+// wrapped total would silently defeat CostTrackerLimit.
+func TestRuntimeCostUnboundedOverloadSaturates(t *testing.T) {
+	unbounded := func(args []ref.Val, result ref.Val) *uint64 {
+		maxCost := uint64(math.MaxUint64)
+		return &maxCost
+	}
+	vars := []*decls.VariableDecl{
+		decls.NewVariable("str1", types.StringType),
+		decls.NewVariable("str2", types.StringType),
+	}
+	in := map[string]any{"str1": "val1", "str2": "val2222222"}
+
+	tests := []struct {
+		name string
+		expr string
+	}{
+		// Each expression accrues cost from the variable references before the unbounded
+		// call is observed, and continues to accrue cost after it in the last two cases.
+		{name: "unbounded call", expr: `"abcdefg".contains(str1 + str2)`},
+		{name: "unbounded call in conjunction", expr: `str1 != "" && "abcdefg".contains(str1 + str2)`},
+		{name: "unbounded call then comparison", expr: `"abcdefg".contains(str1 + str2) == true`},
+		{name: "unbounded call in list", expr: `["abcdefg".contains(str1 + str2), true]`},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			opts := []CostTrackerOption{OverloadCostTracker(overloads.ContainsString, unbounded)}
+			actualCost, _, err := computeCost(t, tc.expr, vars, constructActivation(t, in), opts)
+			if err != nil {
+				t.Fatalf("computeCost() failed: %v", err)
+			}
+			if actualCost != math.MaxUint64 {
+				t.Errorf("computeCost() got cost %d, wanted %d", actualCost, uint64(math.MaxUint64))
+			}
+
+			// The saturated cost must also trip a cost limit.
+			limitOpts := []CostTrackerOption{
+				OverloadCostTracker(overloads.ContainsString, unbounded),
+				CostTrackerLimit(1000),
+			}
+			_, _, err = computeCost(t, tc.expr, vars, constructActivation(t, in), limitOpts)
+			if err == nil {
+				t.Error("computeCost() with a cost limit got nil error, wanted cost limit exceeded")
+			}
+		})
+	}
+}